SARIMA Learner (Labs)

Trains and generates a forecast using a Seasonal AutoRegressive Integrated Moving Average (SARIMA) model. The SARIMA models captures temporal structures in time series data in the following components:

• AR (AutoRegressive): Relationship between the current observation and a number (p) of lagged observations.
• I (Integrated): Degree (d) of differencing required to make the time series stationary.
• MA (Moving Average): Time series mean and the relationship between the current forecast error and a number (q) of lagged forecast errors.

Seasonal versions of these components operate similarly, with lag intervals equal to the seasonal period (S).

Input ports

1. Type: Table
   Input Data

   Table containing training data for fitting the SARIMA model, must contain a numeric target column with no missing values to be used for forecasting.

Output ports

1. Type: Table
   Forecast

   Table containing forecasts for the configured column, the first value will be 1 timestamp ahead of the final training value used.

2. Type: Table
   Residuals

   In sample model prediction values and residuals i.e. difference between observed value and the predicted output.

3. Type: Table
   Coefficients and Statistics

   Table containing fitted model coefficients, variance of residuals (sigma2), and several model metrics along with their standard errors.

4. Type: org.knime.python3.nodes.PythonBinaryBlobFileStorePortObject
   Model

   Pickled model object that can be used by the SARIMA (Apply) node to generate different forecast lengths without refitting the model